OpenAI and Anthropic Open Cyber-Permissive Models (GPT-5.5-Cyber, Mythos) to EU Defenders

OpenAI·medium signal

Under a new EU Cyber Action Plan, OpenAI is opening limited-preview access to GPT-5.5-Cyber — a more permissive variant for vetted defenders covering vulnerability triage, malware analysis, binary reverse engineering, and patch validation — to EU businesses, governments, and institutions including the EU AI Office. Anthropic followed by opening its restricted Mythos model to ENISA days later. OpenAI notes the first cyber preview isn't expected to outperform base GPT-5.5 across every eval; the significance is the geopolitics of gated, defender-only model access.